Deciphering the influence of hydrological variables on water quality via data-driven predictive models.

The water quality of Miyun Reservoir was closely related to that of the rivers flowing into it, the Chao River and Bai River. However, there is a lack of quantitative research on the contributions of rainfall, runoff, sediment, and dissolved oxygen (DO) to water quality of Chao River and Bai River and their optimal ranges. Therefore, this paper used machine learning to predict the impact of these hydrological variables on water quality and optimize these parameters. Results showed that the XGBoost model had the best fitting effect (R2 > 0.9). By multi-objective optimization, the beat range of rainfall, runoff, sediment, and DO were 45.57-174.46 mm, 2.74-4.02 m3/s, 91.7-176.4 kg/s, and 14.1-14.7 mg/L in Chao River, respectively. Meanwhile, the best range of rainfall, runoff, sediment, and DO were 39.8-223.7 mm, 1.14-52 m3/s, 0-151 kg/s, and 7.1-16.3 mg/L in Bai River, respectively. The scenario prediction revealed that when the three basic hydrological indicators of rainfall, runoff, and sediment increased by more than 75% simultaneously, TN at Bai River exceeded the Class III standard of surface water in China first. While at Chao River, not only did TN exceed the Class III, but NH4+-N and TP also exceeded the Class II. In contrast, the Chao River had a significantly higher risk of water quality deterioration under extreme hydrological scenarios than the Bai River, and was more sensitive to hydrological fluctuations. Based on the above results, soil and water conservation work should carried out in the upstream basin of the Chao River and Bai River to improve their water quality. This study provided a comprehensive reference for subsequent manual regulation.